Examples with DownTimeGauge org.apache.flink.runtime.executiongraph.metrics.DownTimeGauge used on opensource projects

Example 1 with DownTimeGauge

use of org.apache.flink.runtime.executiongraph.metrics.DownTimeGauge in project flink by apache.

the class SchedulerBase method registerJobMetrics.

public static void registerJobMetrics(MetricGroup metrics, JobStatusProvider jobStatusProvider, Gauge<Long> numberOfRestarts, DeploymentStateTimeMetrics deploymentTimeMetrics, Consumer<JobStatusListener> jobStatusListenerRegistrar, long initializationTimestamp, MetricOptions.JobStatusMetricsSettings jobStatusMetricsSettings) {
    metrics.gauge(DownTimeGauge.METRIC_NAME, new DownTimeGauge(jobStatusProvider));
    metrics.gauge(UpTimeGauge.METRIC_NAME, new UpTimeGauge(jobStatusProvider));
    metrics.gauge(MetricNames.NUM_RESTARTS, numberOfRestarts);
    metrics.gauge(MetricNames.FULL_RESTARTS, numberOfRestarts);
    final JobStatusMetrics jobStatusMetrics = new JobStatusMetrics(initializationTimestamp, jobStatusMetricsSettings);
    jobStatusMetrics.registerMetrics(metrics);
    jobStatusListenerRegistrar.accept(jobStatusMetrics);
    deploymentTimeMetrics.registerMetrics(metrics);
}

Example 2 with DownTimeGauge

use of org.apache.flink.runtime.executiongraph.metrics.DownTimeGauge in project flink by apache.

the class AdaptiveSchedulerTest method testStatusMetrics.

@Test
public void testStatusMetrics() throws Exception {
    final CompletableFuture<UpTimeGauge> upTimeMetricFuture = new CompletableFuture<>();
    final CompletableFuture<DownTimeGauge> downTimeMetricFuture = new CompletableFuture<>();
    // restartingTime acts as a stand-in for generic status time metrics
    final CompletableFuture<Gauge<Long>> restartTimeMetricFuture = new CompletableFuture<>();
    final MetricRegistry metricRegistry = TestingMetricRegistry.builder().setRegisterConsumer((metric, name, group) -> {
        switch(name) {
            case UpTimeGauge.METRIC_NAME:
                upTimeMetricFuture.complete((UpTimeGauge) metric);
                break;
            case DownTimeGauge.METRIC_NAME:
                downTimeMetricFuture.complete((DownTimeGauge) metric);
                break;
            case "restartingTimeTotal":
                restartTimeMetricFuture.complete((Gauge<Long>) metric);
                break;
        }
    }).build();
    final JobGraph jobGraph = createJobGraph();
    final DefaultDeclarativeSlotPool declarativeSlotPool = createDeclarativeSlotPool(jobGraph.getJobID());
    final Configuration configuration = new Configuration();
    configuration.set(JobManagerOptions.MIN_PARALLELISM_INCREASE, 1);
    configuration.set(JobManagerOptions.RESOURCE_WAIT_TIMEOUT, Duration.ofMillis(10L));
    configuration.set(MetricOptions.JOB_STATUS_METRICS, Arrays.asList(MetricOptions.JobStatusMetrics.TOTAL_TIME));
    final AdaptiveScheduler scheduler = new AdaptiveSchedulerBuilder(jobGraph, singleThreadMainThreadExecutor).setJobMasterConfiguration(configuration).setJobManagerJobMetricGroup(JobManagerMetricGroup.createJobManagerMetricGroup(metricRegistry, "localhost").addJob(new JobID(), "jobName")).setDeclarativeSlotPool(declarativeSlotPool).build();
    final UpTimeGauge upTimeGauge = upTimeMetricFuture.get();
    final DownTimeGauge downTimeGauge = downTimeMetricFuture.get();
    final Gauge<Long> restartTimeGauge = restartTimeMetricFuture.get();
    final SubmissionBufferingTaskManagerGateway taskManagerGateway = new SubmissionBufferingTaskManagerGateway(1 + PARALLELISM);
    taskManagerGateway.setCancelConsumer(createCancelConsumer(scheduler));
    singleThreadMainThreadExecutor.execute(() -> {
        scheduler.startScheduling();
        offerSlots(declarativeSlotPool, createSlotOffersForResourceRequirements(ResourceCounter.withResource(ResourceProfile.UNKNOWN, 1)), taskManagerGateway);
    });
    // wait for the first task submission
    taskManagerGateway.waitForSubmissions(1, Duration.ofSeconds(5));
    // sleep a bit to ensure uptime is > 0
    Thread.sleep(10L);
    assertThat(upTimeGauge.getValue()).isGreaterThan(0L);
    assertThat(downTimeGauge.getValue()).isEqualTo(0L);
    assertThat(restartTimeGauge.getValue()).isEqualTo(0L);
    singleThreadMainThreadExecutor.execute(() -> {
        // offer more slots, which will cause a restart in order to scale up
        offerSlots(declarativeSlotPool, createSlotOffersForResourceRequirements(ResourceCounter.withResource(ResourceProfile.UNKNOWN, 1)), taskManagerGateway);
    });
    // wait for the second task submissions
    taskManagerGateway.waitForSubmissions(2, Duration.ofSeconds(5));
    // sleep a bit to ensure uptime is > 0
    Thread.sleep(10L);
    assertThat(upTimeGauge.getValue()).isGreaterThan(0L);
    assertThat(downTimeGauge.getValue()).isEqualTo(0L);
    // can be zero if the restart is very quick
    assertThat(restartTimeGauge.getValue()).isGreaterThanOrEqualTo(0L);
}